Font Size: a A A

Exploring algorithms for network capacity dimensioning, and traffic modeling along with performance evaluation

Posted on:2004-04-12Degree:Ph.DType:Thesis
University:The University of IowaCandidate:Jiang, HongFull Text:PDF
GTID:2452390011454653Subject:Mathematics
Abstract/Summary:
In this thesis, algorithms for network traffic engineering are explored within three areas. The first area is about network capacity dimensioning, and contains the following four optimization problems: Minimum Cost Routing Problem (MCRP), Balanced Utilization Routing Problem (BURP), Minimum Delay Routing Problem (MDRP) and Minimum Cost Spare Capacity Problem (MCSCP).; The first problem, the MCRP, is solved with the well-developed Lagrangian relaxation algorithm, which takes another efficient algorithm, the “successive approximation push-relabel method” as the procedure to solve the minimum-cost flow problem associated with each commodity in a Lagrangian subproblem.; The remaining three problems are variations of MCRP. Each of them is first converted into one or a series of MCRP, and then solved by an algorithm having the MCRP algorithm as a component. Among these problems, the formulations for BURP and MCSCP are new, and we have not seen similar ones in our literature review.; The second area is about modeling network traffic as self-similar processes, and their parameter estimation. Our main contributions are: proposing a new approximation schema, which is more accurate and more efficient than the existing ones, to calculate the spectral density for FGN (Fractional Gaussian Noise, a special type of self-similar processes); improving the efficiency of the Whittle's estimator (one of the best estimators for self-similarity parameter) significantly by using the proposed approximation schema.; The third area is about algorithms to synthesize self-similar network traffic traces, and performance evaluation by taking the following two types of synthesized traces as input into a simulated queue: self-similar traces with different self-similarity parameter values, and Poisson traces. Our main contributions are: improving the efficiency of a svnthesis algorithm for FGN based on Discrete Time Fourier Transform, by using the approximation schema for the formula of spectral density proposed in the second area; correcting mistakes in a wavelet-based synthesis algorithm and proposing improvements on the accuracy of the algorithm; comparing three performance metrics (packet loss percentage, packet delay, and packet delay jitter) among the different synthesized traces as input into the simulated queue.
Keywords/Search Tags:Algorithm, Network, Traffic, Capacity, Traces, MCRP, Performance, Area
Related items